View Issue Details

Jump to NotesJump to History
IDProjectCategoryView StatusDate SubmittedLast Update
0001450OpenFOAMBugpublic2014-12-02 07:072015-02-25 20:53
Reporteruser1044Assigned Tohenry  
PriorityhighSeveritycrashReproducibilityalways
Status resolvedResolutionfixed 
PlatformLinuxOSOpenSuseOS Version11.2
Summary0001450: decomposePar with scotch breaks for cyclicACMI patch in the tutorial of oscillatingInletACMI2D.
DescriptionI decomposed the tutorial mesh of oscillatingInletACMI2D by scotch, but segmentation fault errors are shown as follows,
/*---------------------------------------------------------------------------*\
| ========= | |
| \\ / F ield | OpenFOAM: The Open Source CFD Toolbox |
| \\ / O peration | Version: 2.3.x |
| \\ / A nd | Web: www.OpenFOAM.org |
| \\/ M anipulation | |
\*---------------------------------------------------------------------------*/
Build : 2.3.x-590d57f32fed
Exec : decomposePar
Date : Dec 02 2014
Time : 16:00:21
Host : "svxxxxx"
PID : 25373
Case : /work3/uxxxxx/OpenFOAM/uxxxxx-2.3.x/tutorials/oscillatingInletACMI2D
nProcs : 1
sigFpe : Enabling floating point exception trapping (FOAM_SIGFPE).
fileModificationChecking : Monitoring run-time modified files using timeStampMaster
allowSystemOperations : Allowing user-supplied system call operations

// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //
Create time



Decomposing mesh region0

Create mesh
AMI: Creating addressing and weights between 40 source faces and 96 target faces
AMI: Patch source sum(weights) min/max/average = 1, 1, 1
AMI: Patch target sum(weights) min/max/average = 0, 1, 0.4

Calculating distribution of cells
Selecting decompositionMethod scotch

Finished decomposition in 0.03 s

Calculating original mesh data

Distributing cells to processors

Distributing faces to processors

Distributing points to processors

Constructing processor meshes
#0 Foam::error::printStack(Foam::Ostream&) in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libOpenFOAM.so"
#1 Foam::sigSegv::sigHandler(int) in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libOpenFOAM.so"
#2 ?? in "/lib64/libc.so.6"
#3 Foam::cyclicACMIPolyPatch::setNeighbourFaceAreas() const in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libmeshTools.so"
#4 Foam::cyclicACMIPolyPatch::resetAMI(Foam::AMIInterpolation<Foam::PrimitivePatch<Foam::face, Foam::SubList, Foam::Field<Foam::Vector<double> > const&, Foam::Vector<double> >, Foam::PrimitivePatch<Foam::face, Foam::SubList, Foam::Field<Foam::Vector<double> > const&, Foam::Vector<double> > >::interpolationMethod const&) const in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libmeshTools.so"
#5 Foam::cyclicACMIPolyPatch::initGeometry(Foam::PstreamBuffers&) in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libmeshTools.so"
#6 Foam::polyBoundaryMesh::calcGeometry() in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libOpenFOAM.so"
#7 Foam::polyMesh::addPatches(Foam::List<Foam::polyPatch*> const&, bool) in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/lib/libOpenFOAM.so"
#8 Foam::domainDecomposition::writeDecomposition(bool) in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/bin/decomposePar"
#9 main in "/home/u15002/OpenFOAM/OpenFOAM-2.3.x/platforms/linux64Gcc48DPOpt/bin/decomposePar"
#10 __libc_start_main in "/lib64/libc.so.6"
#11 _start at /usr/src/packages/BUILD/glibc-2.10.1/csu/../sysdeps/x86_64/elf/start.S:116
Segmentation fault
Steps To Reproduce# cp -r /home/u15002/OpenFOAM/OpenFOAM-2.3.x/tutorials/incompressible/pimpleDyMFoam/oscillatingInletACMI2D/ ./
# cd oscillatingInletACMI2D/
# ./Allrun
Running blockMesh on oscillatingInletACMI2D
Running topoSet on oscillatingInletACMI2D
Running createBaffles on oscillatingInletACMI2D
Running createPatch on oscillatingInletACMI2D
Running pimpleDyMFoam on oscillatingInletACMI2D
^C
# decomposePar
Additional InformationThe number of decomposition with scotch is 4.
TagsACMI, cyclicACMI, decomposePar

Activities

user47

2015-01-19 18:02

  ~0003566

Any update on this issue?

I have exactly the same problem on OpenSUSE 13.2, with the latest git pull of OpenFOAM.

wyldckat

2015-02-14 21:04

updater   ~0003770

Since I had a considerable good look into this, I think it's best to share what I managed to understand.

First of all, a workaround has been reported here: http://www.cfd-online.com/Forums/openfoam-installation/146513-of2-3-1-os13-2-trying-use-dummy-pstream-library.html#post528475 post #19, where temporarily commenting out the method "resetAMI()" in the method "Foam::cyclicACMIPolyPatch::initGeometry" can help bypass the problem for decomposePar.

Attached is the patch "hack_for_bug1450.patch" for somewhat better circumventing this issue, until an improved implementation for ACMI is done. The patch essentially changes in "resetAMI" to only call "setNeighbourFaceAreas()" if there are target faces; if not, it issues a warning message indicating that such a situation should only occur when decomposing a case.


As for the bigger problem, here's what I could figure out:

  1- In decomposePar, treating "cyclicAMIPolyPatch" the same way that "cyclicPolyPatch" is treated, seemed a good idea. It's probably not, taking into account that cyclic patches need strict correlation between the cells that each face corresponds to.

  2- When assuming that "cyclicAMIPolyPatch" should be treated the same way as "cyclicPolyPatch", I triggered the next issue: each cyclicACMI patch relies on two other patches (instead of one), namely the sibling patch and the patch that will be used for covering the unexposed part.

  3- This leads to the problem where we're not able to strictly correlate these 3-4 patches of the same groups to be part of the same processor, simply because the "blockage" patches are standard wall patches, with no coupling identification, apart from the grouping.

  4- This leads me to believe that a new derived patch class would be needed, derived from a wall patch, such as "wallACMI" or something like that, with at least the special "coupled==true" indication and neighbour correlation data. This way we could then at least use the "preservePatches" feature.
 

For now, I guess there isn't much more I can to try and help here, aside from the attached patch file.

wyldckat

2015-02-14 21:06

updater  

hack_for_bug1450.patch (956 bytes)    
diff --git a/src/meshTools/AMIInterpolation/patches/cyclicACMI/cyclicACMIPolyPatch/cyclicACMIPolyPatch.C b/src/meshTools/AMIInterpolation/patches/cyclicACMI/cyclicACMIPolyPatch/cyclicACMIPolyPatch.C
index 46a9468..41ef8de 100644
--- a/src/meshTools/AMIInterpolation/patches/cyclicACMI/cyclicACMIPolyPatch/cyclicACMIPolyPatch.C
+++ b/src/meshTools/AMIInterpolation/patches/cyclicACMI/cyclicACMIPolyPatch/cyclicACMIPolyPatch.C
@@ -98,7 +98,16 @@ void Foam::cyclicACMIPolyPatch::resetAMI
             noSf[faceI] *= 1.0 - srcMask_[faceI];
         }
 
-        setNeighbourFaceAreas();
+        if(tgtMask_.size()>0)
+        {
+          setNeighbourFaceAreas();
+        }
+        else
+        {
+            WarningIn("cyclicAMIPolyPatch::resetAMI(...) const")
+                << "Target mask is not populated." 
+                << " Should only occur when decomposing." << endl;
+        }
 
         // set the updated flag
         updated_ = true;
hack_for_bug1450.patch (956 bytes)   

henry

2015-02-25 20:08

manager   ~0003913

I agree than the cyclicACMIPolyPatch has some SERIOUS issues and will probably need a rewrite but then the cyclicAMIPolyPatch also has SERIOUS issues which need to be addressed first.

In the meantime I am OK with the hack you propose except that is it reasonable to assume that tgtMask_ might be empty indicating that decomposition is taking place? Might tgtMask_ have some elements but a different number to cp.size()?

henry

2015-02-25 20:53

manager   ~0003914

Hacked by commit ea9cfb1213c302d48dd9c55597895e9be9a7fb8b

Issue History

Date Modified Username Field Change
2014-12-02 07:07 user1044 New Issue
2014-12-02 08:53 user1044 Tag Attached: cyclicACMI
2014-12-02 08:53 user1044 Tag Attached: decomposePar
2014-12-29 18:42 wyldckat Tag Attached: ACMI
2015-01-19 18:02 user47 Note Added: 0003566
2015-02-14 21:04 wyldckat Note Added: 0003770
2015-02-14 21:06 wyldckat File Added: hack_for_bug1450.patch
2015-02-25 20:08 henry Note Added: 0003913
2015-02-25 20:53 henry Note Added: 0003914
2015-02-25 20:53 henry Status new => resolved
2015-02-25 20:53 henry Resolution open => fixed
2015-02-25 20:53 henry Assigned To => henry