added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain #10989

mboisson · 2020-07-17T13:47:14Z

There's a missing space at the end of one line of configopts, which means that CMake options get ignored.

mboisson · 2020-07-17T13:47:40Z

@boegel, you missed something there ;)

The current recipe results in -DOPENGL_INCLUDE_DIR=$EBROOTMESA/include-DVTK_OPENGL_HAS_OSMESA=ON in the cmake options.

boegel · 2020-07-19T08:52:12Z

@mboisson Ouch...

Any idea if we can catch this by enhancing the sanity check?

boegel · 2020-07-19T09:35:40Z

Test report by @boegel
FAILED
Build succeeded for 1 out of 2 (2 easyconfigs in this PR)
generoso-x-5 - Linux centos linux 8.2.2004, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/536d329c13f65909153520762d4881a8 for a full test report.

smoors · 2020-07-20T19:17:27Z

Test report by @smoors
FAILED
Build succeeded for 3 out of 4 (2 easyconfigs in this PR)
node381.hydra.os - Linux centos linux 7.7.1908, x86_64, Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz (skylake_avx512), Python 2.7.5
See https://gist.github.com/7b760e6fd4c597de9b5a98c990224900 for a full test report.

mboisson · 2020-07-20T19:19:01Z

Mmmm, so, are those failing tests just a symptoms of the corrected recipe not working, and the incorrect recipe skipping something ?

smoors · 2020-07-20T19:33:16Z

this is the error for the intel version (foss builds fine). looks indeed like an issue with OSMesa linking, which was skipped previously.

[ 65%] Building CXX object CMakeFiles/vtkFiltersPointsCS.dir/CMakeFiles/vtkFiltersPointsCS/vtkLinearKernelClientServer.cxx.o
/apps/brussel/CO7/skylake/software/impi/2019.7.217-iccifort-2020.1.217/intel64/bin/mpiicpc   -I/tmp/3254107.master01.hydra.brussel.vsc/tmp/vsc10009/build/ParaView/5.8.0/intel-2020a-Python-3.8.2-mpi/easybui
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaDestroyContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaMakeCurrent'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaGetCurrentContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaCreateContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaGetProcAddress'
make[2]: *** [bin/vtkProbeOpenGLVersion-pv5.8] Error 1
make[2]: Leaving directory `/tmp/3254107.master01.hydra.brussel.vsc/tmp/vsc10009/build/ParaView/5.8.0/intel-2020a-Python-3.8.2-mpi/easybuild_obj'
make[1]: *** [VTK/Rendering/OpenGL2/CMakeFiles/vtkProbeOpenGLVersion.dir/all] Error 2

boegel · 2020-07-21T09:06:21Z

@boegel one solution would be to allow buildopts and friends be defined a list of strings, which get joined by EB with a space

buildopts can already be a list of string values, but that's used for indicating that the configure-build-test-install cycle myst be run multiple times during a single installation...

smoors · 2020-07-21T12:04:38Z

@boegel one solution would be to allow buildopts and friends be defined a list of strings, which get joined by EB with a space

buildopts can already be a list of string values, but that's used for indicating that the configure-build-test-install cycle myst be run multiple times during a single installation...

ah yes, I knew that..

mboisson · 2020-07-30T15:41:22Z

this is the error for the intel version (foss builds fine). looks indeed like an issue with OSMesa linking, which was skipped previously.

[ 65%] Building CXX object CMakeFiles/vtkFiltersPointsCS.dir/CMakeFiles/vtkFiltersPointsCS/vtkLinearKernelClientServer.cxx.o
/apps/brussel/CO7/skylake/software/impi/2019.7.217-iccifort-2020.1.217/intel64/bin/mpiicpc   -I/tmp/3254107.master01.hydra.brussel.vsc/tmp/vsc10009/build/ParaView/5.8.0/intel-2020a-Python-3.8.2-mpi/easybui
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaDestroyContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaMakeCurrent'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaGetCurrentContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaCreateContext'
ld: ../../../lib64/libvtkRenderingOpenGL2-pv5.8.so.5.8: undefined reference to `OSMesaGetProcAddress'
make[2]: *** [bin/vtkProbeOpenGLVersion-pv5.8] Error 1
make[2]: Leaving directory `/tmp/3254107.master01.hydra.brussel.vsc/tmp/vsc10009/build/ParaView/5.8.0/intel-2020a-Python-3.8.2-mpi/easybuild_obj'
make[1]: *** [VTK/Rendering/OpenGL2/CMakeFiles/vtkProbeOpenGLVersion.dir/all] Error 2

actually, I get this error when building with GCC as well, and even if I use Gentoo's libOSMesa.so.
Looking at the intermediate file, it looks like it does not link at all on libOSMesa.so.

Looks like a ParaView bug to me.

mboisson · 2020-07-30T15:45:16Z

I created an issue on kitware's gitlab... lets see what they say :
https://gitlab.kitware.com/paraview/paraview/-/issues/20120

mboisson · 2020-08-03T15:03:59Z

Related posts :
https://discourse.paraview.org/t/osmesa-linking-problems-compiling-paraview-on-opensuse-leap-15-1/4124/7
https://discourse.paraview.org/t/compiling-paraview-with-vtk-use-x-and-vtk-opengl-has-osmesa-simultaneously/4843

Micket · 2020-08-13T17:59:29Z

I'm here to complicate the situation a bit more: #10977

So, I really really really want a version that works with VirtualGL, which will not work if linked with OSMesa.
I would also like this to be the version without any extra suffixes or such so that it's the primary dependency for OpenFOAM

Then, introduce secondary packages for

ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi-egl.eb
ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi-osmesa.eb

respectively for all your off-screen rendering needs. Probably these versions could just build the pvserver components since this is their usecase.

Since the ParaView we released in the last EB didn't actually build with OSMesa correctly, perhaps we can consider removing it from this PR, and then opening new PRs for a osmesa and egl versions.

mboisson · 2020-08-13T18:05:47Z

I think these recipes actually address what you want to do :
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a-offscreen-gpu.eb
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a-offscreen.eb
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a.eb

We build three flavors of ParaView. One which builds the client, with Qt and Mesa, one which builds fo offscreen rendering with OSMesa, and one which uses EGL from NVidia to work on GPUs.

Micket

I have no strong opinion on any of the other differences, like OSPray and whatnot. But, I would suggest we at least do this.

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi.eb

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-intel-2020a-Python-3.8.2-mpi.eb

boegel · 2020-08-17T12:19:33Z

I'm happy with dropping OSMESA=ON from this Paraview easyconfig, but the implications are not 100% clear to me...

What does enabling OSMESA imply exactly (other than requiring Mesa)?

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi.eb

Micket · 2020-08-17T12:24:57Z

There are basically 3 modes, all, sort-of-mutually-exclusive, rendering backends:

Onscreen, classic GLX, like your classic glxgears (the X in GLX comes from X11)
Offscreen option 1: OSMesa (off-screen-mesa). Not compatible with VirtualGL last i tried (and I can't imagine how it ever would be). Software rendering only. This is a special non-glx part of Mesa.
Offscreen option 2: EGL (the glorious future). Supports hardware rendering off-screen. This is what you really want for pvserver (if you have GPUs). Mesa also supports EGL with software rendering, so, technically, this could superseed everything in the future. But, VirtualGL doesn't support EGL yet, only GLX. So, for Onscreen needs (via VNC, Xpra, ssh, with VirtualGL) you need GLX mode. (b.t.w EGL is what Wayland uses in case someone is curious)

(all 3 of them make use of Mesa)

boegel · 2020-08-17T12:29:19Z

@Micket So that means picking between a Paraview that supports on screen rendering properly (with VirtualGL, no OSMESA) but no offscreen rendering, and one that supports offscreen rendering but no (proper) onscreen rendering?

Don't you basically need both onscreen and offscreen rendering when using Paraview as a dep for OpenFOAM?

Should the Paraview that has good support for on screen rendering be tagged with -gui?

Micket · 2020-08-17T12:55:56Z

@boegel Yes, it means picking. See ComputeCanada's easyconfigs above in @mboisson comments, where they offer all 3 flavours.

OpenFOAM build their own "parafoam" (instead of just writing a custom file format/file-reader plugin, they compile their own paraview 🤡 ) for post-processing.
OpenFOAM definitely isn't doing off-screen rendering during simulations if that was what you where worrying about, it's all post-processing, just like "normal" paraview usage, but with their hacked version to read their input format.
I have always removed the OSMesa stuff in all our ParaView builds locally because VirtualGL support is crucial
ComputeCanada also uses the onscreen flavour as their dep for OpenFOAM.

Some of the versions need to be tagged, yes. My suggestion above was to rather tag the off-screen versions, like ComputeCanada does, though I'm not sure whether it's good to label it "offscreen-gpu" because, technically, EGL supports software rendering as well (though currently one needs to specify extra environment varialbes to coax it into doing so at runtime #10977 (comment) )

But, I don't really care what versionsuffixes. EGL support for VirtualGL is getting closer and closer, and with some simple patches to VTK, it should definitely be possible for EGL to superseed everything; onscreen and offscreen, gpu and software rendering, so, eventually we could just have 1 golden version again.

bartoldeman · 2020-08-17T14:41:19Z

In general we instruct direct users of paraview to install its client on their own laptop/workstation/desktop/whatever and then use the server, on the cluster which then will either use a GPU or software rendering, see https://docs.computecanada.ca/wiki/ParaView

For the rest I agree with @Micket here. For us running paraview via virtualgl is a secondary use case but that's just a local detail.

boegel · 2021-01-08T13:14:18Z

@Micket @mboisson I would like to get this merged ASAP...

What are the next steps here? The Mesa dependency needs to stay, right? So it's only a matter of test reports & merging?

Micket · 2021-01-08T14:04:09Z

No matter what we do, we need to have Mesa as a dependency. Mesa supplies the libraries for all 3 possible backends (good old GLX, the obscure "OSMESA" and the new-fangled EGL).

I suggest we

Drop OSMESA from this package, and consider this the on-screen version.
Introduce a new easyconfig with OSMESA support and add a suffix -offscreen
Introduce yet another easyconfig with EGL support and add a suffix -offscreen-gpu

(I have no particular strong feelings about what the suffix should be, just something along these lines)

mboisson · 2021-01-08T14:22:25Z

No matter what we do, we need to have Mesa as a dependency. Mesa supplies the libraries for all 3 possible backends (good old GLX, the obscure "OSMESA" and the new-fangled EGL).

I suggest we

Drop OSMESA from this package, and consider this the on-screen version.

Introduce a new easyconfig with OSMESA support and add a suffix -offscreen

Introduce yet another easyconfig with EGL support and add a suffix -offscreen-gpu

(I have no particular strong feelings about what the suffix should be, just something along these lines)

These would fit exactly what we do in Compute Canada, so I'm fine with those. For information, these are what we have installed:
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a.eb
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a-offscreen.eb
https://github.com/ComputeCanada/easybuild-easyconfigs/blob/computecanada-master/easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-gompi-2020a-offscreen-gpu.eb

Micket · 2021-01-08T14:29:29Z

If there aren't any objections, then I suggest we do this suggested change (which is what we were implicitly already doing here due to the typo) #10989 (review) ?

boegel · 2021-01-08T14:34:32Z

If there aren't any objections, then I suggest we do this suggested change (which is what we were implicitly already doing here due to the typo) #10989 (review) ?

Agreed.

@mboisson Can you also remove the -DVTK_OPENGL_HAS_OSMESA=ON?

boegel

lgtm

boegel · 2021-01-08T16:44:11Z

@boegelbot please test @ generoso

boegelbot · 2021-01-08T16:45:30Z

@boegel: Request for testing this PR well received on generoso

PR test command 'EB_PR=10989 EB_ARGS= /apps/slurm/default/bin/sbatch --job-name test_PR_10989 --ntasks=4 ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

exit code: 0
output:

Submitted batch job 12398

Test results coming soon (I hope)...

- notification for comment with ID 756862528 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

Micket

lgtm

boegelbot · 2021-01-08T19:18:51Z

Test report by @boegelbot
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in total)
generoso-x-3 - Linux centos linux 8.2.2004, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/b8e9d1c7a947f992d9f45040f15d67e8 for a full test report.

boegel · 2021-01-08T22:04:30Z

Test report by @boegel
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in total)
node3504.doduo.os - Linux RHEL 8.2, x86_64, AMD EPYC 7552 48-Core Processor (zen2), Python 3.6.8
See https://gist.github.com/3dfb09d8d8a86ff91f3fa6994e194ce0 for a full test report.

boegel · 2021-01-08T22:20:28Z

Going in, thanks @mboisson!

boegel · 2021-01-09T02:56:47Z

Test report by @boegel
SUCCESS
Build succeeded for 2 out of 2 (2 easyconfigs in total)
node2659.swalot.os - Linux centos linux 7.9.2009, x86_64, Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz (haswell), Python 3.6.8
See https://gist.github.com/bb70488812dc41a8f4fa036b39b57ff7 for a full test report.

added missing space

14e94f4

boegel added the bug fix label Jul 19, 2020

boegel added this to the next release (4.2.3?) milestone Jul 19, 2020

boegel changed the title ~~added missing space~~ added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain Jul 19, 2020

Micket requested changes Aug 13, 2020

View reviewed changes

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi.eb Outdated Show resolved Hide resolved

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-intel-2020a-Python-3.8.2-mpi.eb Outdated Show resolved Hide resolved

boegel requested changes Aug 17, 2020

View reviewed changes

easybuild/easyconfigs/p/ParaView/ParaView-5.8.0-foss-2020a-Python-3.8.2-mpi.eb Outdated Show resolved Hide resolved

boegel modified the milestones: next release (4.3.0), release after 4.3.0 (4.3.1?) Sep 9, 2020

boegel modified the milestones: 4.3.1 (next release), 4.3.2 Oct 26, 2020

boegel modified the milestones: 4.3.2 (next release), 4.4.0 Dec 8, 2020

boegel mentioned this pull request Jan 8, 2021

{cae}[foss/2020b] OpenFOAM v8, ParaView v5.8.1, CGAL v5.2 #11970

Merged

Merge branch 'upstream_develop' into fix_paraview

03c4065

mboisson added 2 commits January 8, 2021 14:39

removed OSMESA

be59f78

removed OSMESA from the intel version as well

96bdaa1

boegel approved these changes Jan 8, 2021

View reviewed changes

Micket approved these changes Jan 8, 2021

View reviewed changes

boegel merged commit 8f6ba5f into easybuilders:develop Jan 8, 2021

mboisson deleted the fix_paraview branch March 26, 2021 18:45

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain #10989

added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain #10989

mboisson commented Jul 17, 2020

mboisson commented Jul 17, 2020 •

edited

Loading

boegel commented Jul 19, 2020

boegel commented Jul 19, 2020

smoors commented Jul 20, 2020

mboisson commented Jul 20, 2020

smoors commented Jul 20, 2020

boegel commented Jul 21, 2020

smoors commented Jul 21, 2020

mboisson commented Jul 30, 2020

mboisson commented Jul 30, 2020 •

edited

Loading

mboisson commented Aug 3, 2020

Micket commented Aug 13, 2020

mboisson commented Aug 13, 2020

Micket left a comment

boegel commented Aug 17, 2020

Micket commented Aug 17, 2020 •

edited

Loading

boegel commented Aug 17, 2020

Micket commented Aug 17, 2020 •

edited

Loading

bartoldeman commented Aug 17, 2020

boegel commented Jan 8, 2021

Micket commented Jan 8, 2021

mboisson commented Jan 8, 2021

Micket commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel left a comment

boegel commented Jan 8, 2021

boegelbot commented Jan 8, 2021

Micket left a comment

boegelbot commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel commented Jan 9, 2021

added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain #10989

added missing space in 'configopts' in ParaView 5.8.0 easyconfigs using 2020a toolchain #10989

Conversation

mboisson commented Jul 17, 2020

mboisson commented Jul 17, 2020 • edited Loading

boegel commented Jul 19, 2020

boegel commented Jul 19, 2020

smoors commented Jul 20, 2020

mboisson commented Jul 20, 2020

smoors commented Jul 20, 2020

boegel commented Jul 21, 2020

smoors commented Jul 21, 2020

mboisson commented Jul 30, 2020

mboisson commented Jul 30, 2020 • edited Loading

mboisson commented Aug 3, 2020

Micket commented Aug 13, 2020

mboisson commented Aug 13, 2020

Micket left a comment

Choose a reason for hiding this comment

boegel commented Aug 17, 2020

Micket commented Aug 17, 2020 • edited Loading

boegel commented Aug 17, 2020

Micket commented Aug 17, 2020 • edited Loading

bartoldeman commented Aug 17, 2020

boegel commented Jan 8, 2021

Micket commented Jan 8, 2021

mboisson commented Jan 8, 2021

Micket commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel left a comment

Choose a reason for hiding this comment

boegel commented Jan 8, 2021

boegelbot commented Jan 8, 2021

Micket left a comment

Choose a reason for hiding this comment

boegelbot commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel commented Jan 8, 2021

boegel commented Jan 9, 2021

mboisson commented Jul 17, 2020 •

edited

Loading

mboisson commented Jul 30, 2020 •

edited

Loading

Micket commented Aug 17, 2020 •

edited

Loading

Micket commented Aug 17, 2020 •

edited

Loading