[GRASS-user] r.sun.daily with multiple CPU cores: error uncompressing raster data ...

[Dylan Beaudette]

Dang, this appears to be happening with any module run in parallel
that generates FCELL or DCELL maps. In this case, r.horizon run in
parallel.

I have added additional commentary to #2764

https://trac.osgeo.org/grass/ticket/2764#comment:9

After application of Markus' latest patch the error message is now
"Decompression failed with error 0".

Dylan

On Sat, Dec 30, 2017 at 9:34 PM, Dylan Beaudette
<dylan.beaudette at gmail.com> wrote:
> On Sat, Dec 30, 2017 at 2:32 PM, Markus Metz
> <markus.metz.giswork at gmail.com> wrote:
>>
>>
>> On Fri, Dec 29, 2017 at 5:07 PM, Dylan Beaudette <dylan.beaudette at gmail.com>
>> wrote:
>>>
>>> Hi everyone,
>>>
>>> First of all, thanks in advance for the r.sun.daily module which is a
>>> nice replacement for my amateurish attempts over the last 12 years.
>>>
>>> I am currently working on an annual beam radiance map for a large
>>> geographic region, at 30m res: 70,953 x 46,964 cells. This is far too
>>> large for a single pass of r.horizon or r.sun on my machine so I have
>>> split the data into 5,000 x 5,000 cell tiles with 100 cells of
>>> overlap. This seems to be sufficient for my purposes and the edge
>>> effects are tolerable.
>>>
>>> At 8-15 minutes / tile / day (r.sun) and 54 tiles this job calls for
>>> multiple CPU cores. All of the parallel processing that I use is (as
>>> far as I know) contained within the same region.
>>>
>>>
>>> I have had good success with running r.horizon in parallel via GNU
>>> parallel like this:
>>>
>>> # 1: start angle
>>> # 2: angle step
>>> # 3: elevation tile
>>> seq 0 $step 330 | parallel --gnu --progress "bash make-hz-maps.sh {}
>>> $step $elev"
>>>
>>> Which is just a wrapper around r.horizon and run in parallel "within"
>>> tiles.
>>>
>>> Next, I run r.sun.daily (8 CPU cores) within tiles:
>>>
>>> r.sun.daily --overwrite elevation=$elev aspect=$asp slope=$slope \
>>> start_day=1 end_day=365 beam_rad=$beam horizon_basename=hzangle
>>> horizon_step=$step nprocs=8
>>>
>>>
>>> The r.sun.daily modules finishes without error about 50-60% of the
>>> time, results look good. The other 50-40% of the the time I get this:
>>>
>>>
>>> ERROR: Error uncompressing raster data for row 2605 of
>>>        <r_sun_crop8255_beam_rad_tmp_352>
>>> *** buffer overflow detected ***: g.message terminated
>>> ======= Backtrace: =========
>>> /lib/x86_64-linux-gnu/libc.so.6(+0x7329f)[0x7f8b9a70129f]
>>> /lib/x86_64-linux-gnu/libc.so.6(__fortify_fail+0x5c)[0x7f8b9a79c83c]
>>> /lib/x86_64-linux-gnu/libc.so.6(+0x10d710)[0x7f8b9a79b710]
>>> /lib/x86_64-linux-gnu/libc.so.6(+0x10cc19)[0x7f8b9a79ac19]
>>> /lib/x86_64-linux-gnu/libc.so.6(_IO_default_xsputn+0xbc)[0x7f8b9a70961c]
>>> /lib/x86_64-linux-gnu/libc.so.6(_IO_vfprintf+0x1cc5)[0x7f8b9a6d9905]
>>> /lib/x86_64-linux-gnu/libc.so.6(__vsprintf_chk+0x84)[0x7f8b9a79aca4]
>>>
>>> /usr/local/grass-7.5.svn/lib/libgrass_gis.7.5.svn.so(+0x1343c)[0x7f8b9aa6a43c]
>>>
>>> /usr/local/grass-7.5.svn/lib/libgrass_gis.7.5.svn.so(G_fatal_error+0xbf)[0x7f8b9aa6accf]
>>> g.message(main+0x254)[0x400dd4]
>>> /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf5)[0x7f8b9a6aff45]
>>> g.message[0x400ea2]
>>>
>>>
>>> I can't seem to replicate the problem, as subsequent runs with the
>>> same parameters and in the same tile are successful! This leads me to
>>> think that:
>>>
>>> * some aspect of this approach is not thread safe
>>> * there is something wrong with my computer
>>> * there is a subtle bug in the raster writing / reading code when
>>> invoked in parallel
>>>
>>>
>>> I have encountered similar raster reading errors in the past,
>>> typically in the context of parallel processing:
>>>
>>> https://trac.osgeo.org/grass/ticket/2762
>>>
>>> https://trac.osgeo.org/grass/ticket/2764
>>>
>>>
>>> http://osgeo-org.1560.x6.nabble.com/Error-reading-raster-data-for-row-xxx-only-when-using-r-series-and-t-rast-series-td5229569i20.html
>>>
>>> https://lists.osgeo.org/pipermail/grass-dev/2015-July/075691.html
>>>
>>>
>>> Here is some information on my system and version of GRASS:
>>>
>>>  ./configure  --enable-64bit --with-libs=/usr/lib --without-pthread
>>> --without-odbc --without-mysql --with-readline --with-cxx
>>> --enable-largefile --with-freetype
>>> --with-freetype-includes=/usr/include/freetype2 --with-sqlite
>>> --with-python --with-geos=/usr/local/bin/geos-config --without-opencl
>>> --with-opencl-includes=/usr/include/CL/ --with-postgres
>>> --with-postgres-includes=/usr/include/postgresql/
>>> --with-postgres-libs=/usr/lib/
>>> --with-proj-share=/usr/local/share/proj/
>>>
>>> version=7.5.svn
>>> date=2017
>>> revision=r71964
>>> build_date=2017-12-21
>>> build_platform=x86_64-pc-linux-gnu
>>> build_off_t_size=8
>>>
>>>
>>> Any ideas?
>>
>> Please try the patch attached to ticket #2764 helps to get closer to the
>> problem.
>>
>> Markus M
>>
>
> Hi Markus,
>
> Thank you for the quick reply and patch. I have recompiled with the
> patch from #2764 and now waiting to see what happens.
>
> Happy New Year!
>
> Dylan