Files
OpenFOAM-12/applications/utilities/parallelProcessing/reconstructParMesh/reconstructParMesh.C
Henry Weller 7c301dbff4 Parallel IO: New collated file format
When an OpenFOAM simulation runs in parallel, the data for decomposed fields and
mesh(es) has historically been stored in multiple files within separate
directories for each processor.  Processor directories are named 'processorN',
where N is the processor number.

This commit introduces an alternative "collated" file format where the data for
each decomposed field (and mesh) is collated into a single file, which is
written and read on the master processor.  The files are stored in a single
directory named 'processors'.

The new format produces significantly fewer files - one per field, instead of N
per field.  For large parallel cases, this avoids the restriction on the number
of open files imposed by the operating system limits.

The file writing can be threaded allowing the simulation to continue running
while the data is being written to file.  NFS (Network File System) is not
needed when using the the collated format and additionally, there is an option
to run without NFS with the original uncollated approach, known as
"masterUncollated".

The controls for the file handling are in the OptimisationSwitches of
etc/controlDict:

OptimisationSwitches
{
    ...

    //- Parallel IO file handler
    //  uncollated (default), collated or masterUncollated
    fileHandler uncollated;

    //- collated: thread buffer size for queued file writes.
    //  If set to 0 or not sufficient for the file size threading is not used.
    //  Default: 2e9
    maxThreadFileBufferSize 2e9;

    //- masterUncollated: non-blocking buffer size.
    //  If the file exceeds this buffer size scheduled transfer is used.
    //  Default: 2e9
    maxMasterFileBufferSize 2e9;
}

When using the collated file handling, memory is allocated for the data in the
thread.  maxThreadFileBufferSize sets the maximum size of memory in bytes that
is allocated.  If the data exceeds this size, the write does not use threading.

When using the masterUncollated file handling, non-blocking MPI communication
requires a sufficiently large memory buffer on the master node.
maxMasterFileBufferSize sets the maximum size in bytes of the buffer.  If the
data exceeds this size, the system uses scheduled communication.

The installation defaults for the fileHandler choice, maxThreadFileBufferSize
and maxMasterFileBufferSize (set in etc/controlDict) can be over-ridden within
the case controlDict file, like other parameters.  Additionally the fileHandler
can be set by:
- the "-fileHandler" command line argument;
- a FOAM_FILEHANDLER environment variable.

A foamFormatConvert utility allows users to convert files between the collated
and uncollated formats, e.g.
    mpirun -np 2 foamFormatConvert -parallel -fileHandler uncollated

An example case demonstrating the file handling methods is provided in:
$FOAM_TUTORIALS/IO/fileHandling

The work was undertaken by Mattijs Janssens, in collaboration with Henry Weller.
2017-07-07 11:39:56 +01:00

1020 lines
30 KiB
C

/*---------------------------------------------------------------------------*\
========= |
\\ / F ield | OpenFOAM: The Open Source CFD Toolbox
\\ / O peration |
\\ / A nd | Copyright (C) 2011-2017 OpenFOAM Foundation
\\/ M anipulation |
-------------------------------------------------------------------------------
License
This file is part of OpenFOAM.
OpenFOAM is free software: you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
OpenFOAM is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License
along with OpenFOAM. If not, see <http://www.gnu.org/licenses/>.
Application
reconstructParMesh
Description
Reconstructs a mesh using geometric information only.
Writes point/face/cell procAddressing so afterwards reconstructPar can be
used to reconstruct fields.
Note:
- uses geometric matching tolerance (set with -mergeTol (at your option)
If the parallel case does not have correct procBoundaries use the
-fullMatch option which will check all boundary faces (bit slower).
\*---------------------------------------------------------------------------*/
#include "argList.H"
#include "timeSelector.H"
#include "IOobjectList.H"
#include "labelIOList.H"
#include "processorPolyPatch.H"
#include "mapAddedPolyMesh.H"
#include "polyMeshAdder.H"
#include "faceCoupleInfo.H"
#include "fvMeshAdder.H"
#include "polyTopoChange.H"
#include "extrapolatedCalculatedFvPatchFields.H"
using namespace Foam;
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //
// Tolerance (as fraction of the bounding box). Needs to be fairly lax since
// usually meshes get written with limited precision (6 digits)
static const scalar defaultMergeTol = 1e-7;
static void renumber
(
const labelList& map,
labelList& elems
)
{
forAll(elems, i)
{
if (elems[i] >= 0)
{
elems[i] = map[elems[i]];
}
}
}
// Determine which faces are coupled. Uses geometric merge distance.
// Looks either at all boundaryFaces (fullMatch) or only at the
// procBoundaries for proci. Assumes that masterMesh contains already merged
// all the processors < proci.
autoPtr<faceCoupleInfo> determineCoupledFaces
(
const bool fullMatch,
const label masterMeshProcStart,
const label masterMeshProcEnd,
const polyMesh& masterMesh,
const label meshToAddProcStart,
const label meshToAddProcEnd,
const polyMesh& meshToAdd,
const scalar mergeDist
)
{
if (fullMatch || masterMesh.nCells() == 0)
{
return autoPtr<faceCoupleInfo>
(
new faceCoupleInfo
(
masterMesh,
meshToAdd,
mergeDist, // Absolute merging distance
true // Matching faces identical
)
);
}
else
{
// Pick up all patches on masterMesh ending in "toDDD" where DDD is
// the processor number proci.
const polyBoundaryMesh& masterPatches = masterMesh.boundaryMesh();
DynamicList<label> masterFaces
(
masterMesh.nFaces()
- masterMesh.nInternalFaces()
);
forAll(masterPatches, patchi)
{
const polyPatch& pp = masterPatches[patchi];
if (isA<processorPolyPatch>(pp))
{
for
(
label proci=meshToAddProcStart;
proci<meshToAddProcEnd;
proci++
)
{
const string toProcString("to" + name(proci));
if (
pp.name().rfind(toProcString)
== (pp.name().size()-toProcString.size())
)
{
label meshFacei = pp.start();
forAll(pp, i)
{
masterFaces.append(meshFacei++);
}
break;
}
}
}
}
masterFaces.shrink();
// Pick up all patches on meshToAdd ending in "procBoundaryDDDtoYYY"
// where DDD is the processor number proci and YYY is < proci.
const polyBoundaryMesh& addPatches = meshToAdd.boundaryMesh();
DynamicList<label> addFaces
(
meshToAdd.nFaces()
- meshToAdd.nInternalFaces()
);
forAll(addPatches, patchi)
{
const polyPatch& pp = addPatches[patchi];
if (isA<processorPolyPatch>(pp))
{
bool isConnected = false;
for
(
label mergedProci=masterMeshProcStart;
!isConnected && (mergedProci < masterMeshProcEnd);
mergedProci++
)
{
for
(
label proci = meshToAddProcStart;
proci < meshToAddProcEnd;
proci++
)
{
const word fromProcString
(
processorPolyPatch::newName(proci, mergedProci)
);
if (pp.name() == fromProcString)
{
isConnected = true;
break;
}
}
}
if (isConnected)
{
label meshFacei = pp.start();
forAll(pp, i)
{
addFaces.append(meshFacei++);
}
}
}
}
addFaces.shrink();
return autoPtr<faceCoupleInfo>
(
new faceCoupleInfo
(
masterMesh,
masterFaces,
meshToAdd,
addFaces,
mergeDist, // Absolute merging distance
true, // Matching faces identical?
false, // If perfect match are faces already ordered
// (e.g. processor patches)
false // are faces each on separate patch?
)
);
}
}
autoPtr<mapPolyMesh> mergeSharedPoints
(
const scalar mergeDist,
polyMesh& mesh,
labelListList& pointProcAddressing
)
{
// Find out which sets of points get merged and create a map from
// mesh point to unique point.
Map<label> pointToMaster
(
fvMeshAdder::findSharedPoints
(
mesh,
mergeDist
)
);
Info<< "mergeSharedPoints : detected " << pointToMaster.size()
<< " points that are to be merged." << endl;
if (returnReduce(pointToMaster.size(), sumOp<label>()) == 0)
{
return autoPtr<mapPolyMesh>(nullptr);
}
polyTopoChange meshMod(mesh);
fvMeshAdder::mergePoints(mesh, pointToMaster, meshMod);
// Change the mesh (no inflation). Note: parallel comms allowed.
autoPtr<mapPolyMesh> map = meshMod.changeMesh(mesh, false, true);
// Update fields. No inflation, parallel sync.
mesh.updateMesh(map);
// pointProcAddressing give indices into the master mesh so adapt them
// for changed point numbering.
// Adapt constructMaps for merged points.
forAll(pointProcAddressing, proci)
{
labelList& constructMap = pointProcAddressing[proci];
forAll(constructMap, i)
{
label oldPointi = constructMap[i];
// New label of point after changeMesh.
label newPointi = map().reversePointMap()[oldPointi];
if (newPointi < -1)
{
constructMap[i] = -newPointi-2;
}
else if (newPointi >= 0)
{
constructMap[i] = newPointi;
}
else
{
FatalErrorInFunction
<< "Problem. oldPointi:" << oldPointi
<< " newPointi:" << newPointi << abort(FatalError);
}
}
}
return map;
}
boundBox procBounds
(
const argList& args,
const PtrList<Time>& databases,
const word& regionDir
)
{
boundBox bb = boundBox::invertedBox;
forAll(databases, proci)
{
fileName pointsInstance
(
databases[proci].findInstance
(
regionDir/polyMesh::meshSubDir,
"points"
)
);
if (pointsInstance != databases[proci].timeName())
{
FatalErrorInFunction
<< "Your time was specified as " << databases[proci].timeName()
<< " but there is no polyMesh/points in that time." << endl
<< "(there is a points file in " << pointsInstance
<< ")" << endl
<< "Please rerun with the correct time specified"
<< " (through the -constant, -time or -latestTime "
<< "(at your option)."
<< endl << exit(FatalError);
}
Info<< "Reading points from "
<< databases[proci].caseName()
<< " for time = " << databases[proci].timeName()
<< nl << endl;
pointIOField points
(
IOobject
(
"points",
databases[proci].findInstance
(
regionDir/polyMesh::meshSubDir,
"points"
),
regionDir/polyMesh::meshSubDir,
databases[proci],
IOobject::MUST_READ,
IOobject::NO_WRITE,
false
)
);
boundBox domainBb(points, false);
bb.min() = min(bb.min(), domainBb.min());
bb.max() = max(bb.max(), domainBb.max());
}
return bb;
}
void writeCellDistance
(
Time& runTime,
const fvMesh& masterMesh,
const labelListList& cellProcAddressing
)
{
// Write the decomposition as labelList for use with 'manual'
// decomposition method.
labelIOList cellDecomposition
(
IOobject
(
"cellDecomposition",
masterMesh.facesInstance(),
masterMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false
),
masterMesh.nCells()
);
forAll(cellProcAddressing, proci)
{
const labelList& pCells = cellProcAddressing[proci];
UIndirectList<label>(cellDecomposition, pCells) = proci;
}
cellDecomposition.write();
Info<< nl << "Wrote decomposition to "
<< cellDecomposition.objectPath()
<< " for use in manual decomposition." << endl;
// Write as volScalarField for postprocessing. Change time to 0
// if was 'constant'
{
const scalar oldTime = runTime.value();
const label oldIndex = runTime.timeIndex();
if (runTime.timeName() == runTime.constant() && oldIndex == 0)
{
runTime.setTime(0, oldIndex+1);
}
volScalarField cellDist
(
IOobject
(
"cellDist",
runTime.timeName(),
masterMesh,
IOobject::NO_READ,
IOobject::AUTO_WRITE
),
masterMesh,
dimensionedScalar("cellDist", dimless, 0),
extrapolatedCalculatedFvPatchScalarField::typeName
);
forAll(cellDecomposition, celli)
{
cellDist[celli] = cellDecomposition[celli];
}
cellDist.correctBoundaryConditions();
cellDist.write();
Info<< nl << "Wrote decomposition as volScalarField to "
<< cellDist.name() << " for use in postprocessing."
<< endl;
// Restore time
runTime.setTime(oldTime, oldIndex);
}
}
int main(int argc, char *argv[])
{
argList::addNote
(
"reconstruct a mesh using geometric information only"
);
// Enable -constant ... if someone really wants it
// Enable -withZero to prevent accidentally trashing the initial fields
timeSelector::addOptions(true, true);
argList::noParallel();
argList::addOption
(
"mergeTol",
"scalar",
"specify the merge distance relative to the bounding box size "
"(default 1e-7)"
);
argList::addBoolOption
(
"fullMatch",
"do (slower) geometric matching on all boundary faces"
);
argList::addBoolOption
(
"cellDist",
"write cell distribution as a labelList - for use with 'manual' "
"decomposition method or as a volScalarField for post-processing."
);
#include "addRegionOption.H"
#include "setRootCase.H"
#include "createTime.H"
Info<< "This is an experimental tool which tries to merge"
<< " individual processor" << nl
<< "meshes back into one master mesh. Use it if the original"
<< " master mesh has" << nl
<< "been deleted or if the processor meshes have been modified"
<< " (topology change)." << nl
<< "This tool will write the resulting mesh to a new time step"
<< " and construct" << nl
<< "xxxxProcAddressing files in the processor meshes so"
<< " reconstructPar can be" << nl
<< "used to regenerate the fields on the master mesh." << nl
<< nl
<< "Not well tested & use at your own risk!" << nl
<< endl;
word regionName = polyMesh::defaultRegion;
word regionDir = word::null;
if
(
args.optionReadIfPresent("region", regionName)
&& regionName != polyMesh::defaultRegion
)
{
regionDir = regionName;
Info<< "Operating on region " << regionName << nl << endl;
}
scalar mergeTol = defaultMergeTol;
args.optionReadIfPresent("mergeTol", mergeTol);
scalar writeTol = Foam::pow(10.0, -scalar(IOstream::defaultPrecision()));
Info<< "Merge tolerance : " << mergeTol << nl
<< "Write tolerance : " << writeTol << endl;
if (runTime.writeFormat() == IOstream::ASCII && mergeTol < writeTol)
{
FatalErrorInFunction
<< "Your current settings specify ASCII writing with "
<< IOstream::defaultPrecision() << " digits precision." << endl
<< "Your merging tolerance (" << mergeTol << ") is finer than this."
<< endl
<< "Please change your writeFormat to binary"
<< " or increase the writePrecision" << endl
<< "or adjust the merge tolerance (-mergeTol)."
<< exit(FatalError);
}
const bool fullMatch = args.optionFound("fullMatch");
if (fullMatch)
{
Info<< "Doing geometric matching on all boundary faces." << nl << endl;
}
else
{
Info<< "Doing geometric matching on correct procBoundaries only."
<< nl << "This assumes a correct decomposition." << endl;
}
bool writeCellDist = args.optionFound("cellDist");
label nProcs = fileHandler().nProcs(args.path());
Info<< "Found " << nProcs << " processor directories" << nl << endl;
// Read all time databases
PtrList<Time> databases(nProcs);
forAll(databases, proci)
{
Info<< "Reading database "
<< args.caseName()/fileName(word("processor") + name(proci))
<< endl;
databases.set
(
proci,
new Time
(
Time::controlDictName,
args.rootPath(),
args.caseName()/fileName(word("processor") + name(proci))
)
);
}
// Use the times list from the master processor
// and select a subset based on the command-line options
instantList timeDirs = timeSelector::select
(
databases[0].times(),
args
);
// Loop over all times
forAll(timeDirs, timeI)
{
// Set time for global database
runTime.setTime(timeDirs[timeI], timeI);
Info<< "Time = " << runTime.timeName() << nl << endl;
// Set time for all databases
forAll(databases, proci)
{
databases[proci].setTime(timeDirs[timeI], timeI);
}
IOobject facesIO
(
"faces",
databases[0].timeName(),
regionDir/polyMesh::meshSubDir,
databases[0],
IOobject::NO_READ,
IOobject::NO_WRITE
);
// Problem: faceCompactIOList recognises both 'faceList' and
// 'faceCompactList' so we should be lenient when doing
// typeHeaderOk
if (!facesIO.typeHeaderOk<faceCompactIOList>(false))
{
Info<< "No mesh." << nl << endl;
continue;
}
// Read point on individual processors to determine merge tolerance
// (otherwise single cell domains might give problems)
const boundBox bb = procBounds(args, databases, regionDir);
const scalar mergeDist = mergeTol*bb.mag();
Info<< "Overall mesh bounding box : " << bb << nl
<< "Relative tolerance : " << mergeTol << nl
<< "Absolute matching distance : " << mergeDist << nl
<< endl;
// Addressing from processor to reconstructed case
labelListList cellProcAddressing(nProcs);
labelListList faceProcAddressing(nProcs);
labelListList pointProcAddressing(nProcs);
labelListList boundaryProcAddressing(nProcs);
// Internal faces on the final reconstructed mesh
label masterInternalFaces;
// Owner addressing on the final reconstructed mesh
labelList masterOwner;
{
// Construct empty mesh.
// fvMesh** masterMesh = new fvMesh*[nProcs];
PtrList<fvMesh> masterMesh(nProcs);
for (label proci=0; proci<nProcs; proci++)
{
masterMesh.set
(
proci,
new fvMesh
(
IOobject
(
regionName,
runTime.timeName(),
runTime,
IOobject::NO_READ
),
xferCopy(pointField()),
xferCopy(faceList()),
xferCopy(cellList())
)
);
fvMesh meshToAdd
(
IOobject
(
regionName,
databases[proci].timeName(),
databases[proci]
)
);
// Initialize its addressing
cellProcAddressing[proci] = identity(meshToAdd.nCells());
faceProcAddressing[proci] = identity(meshToAdd.nFaces());
pointProcAddressing[proci] = identity(meshToAdd.nPoints());
boundaryProcAddressing[proci] =
identity(meshToAdd.boundaryMesh().size());
// Find geometrically shared points/faces.
autoPtr<faceCoupleInfo> couples = determineCoupledFaces
(
fullMatch,
proci,
proci,
masterMesh[proci],
proci,
proci,
meshToAdd,
mergeDist
);
// Add elements to mesh
autoPtr<mapAddedPolyMesh> map = fvMeshAdder::add
(
masterMesh[proci],
meshToAdd,
couples
);
// Added processor
renumber(map().addedCellMap(), cellProcAddressing[proci]);
renumber(map().addedFaceMap(), faceProcAddressing[proci]);
renumber(map().addedPointMap(), pointProcAddressing[proci]);
renumber(map().addedPatchMap(), boundaryProcAddressing[proci]);
}
for (label step=2; step<nProcs*2; step*=2)
{
for (label proci=0; proci<nProcs; proci+=step)
{
label next = proci + step/2;
if(next >= nProcs)
{
continue;
}
Info<< "Merging mesh " << proci << " with " << next << endl;
// Find geometrically shared points/faces.
autoPtr<faceCoupleInfo> couples = determineCoupledFaces
(
fullMatch,
proci,
next,
masterMesh[proci],
next,
proci+step,
masterMesh[next],
mergeDist
);
// Add elements to mesh
autoPtr<mapAddedPolyMesh> map = fvMeshAdder::add
(
masterMesh[proci],
masterMesh[next],
couples
);
// Processors that were already in masterMesh
for (label mergedI=proci; mergedI<next; mergedI++)
{
renumber
(
map().oldCellMap(),
cellProcAddressing[mergedI]
);
renumber
(
map().oldFaceMap(),
faceProcAddressing[mergedI]
);
renumber
(
map().oldPointMap(),
pointProcAddressing[mergedI]
);
// Note: boundary is special since can contain -1.
renumber
(
map().oldPatchMap(),
boundaryProcAddressing[mergedI]
);
}
// Added processor
for
(
label addedI=next;
addedI<min(proci+step, nProcs);
addedI++
)
{
renumber
(
map().addedCellMap(),
cellProcAddressing[addedI]
);
renumber
(
map().addedFaceMap(),
faceProcAddressing[addedI]
);
renumber
(
map().addedPointMap(),
pointProcAddressing[addedI]
);
renumber
(
map().addedPatchMap(),
boundaryProcAddressing[addedI]
);
}
masterMesh.set(next, nullptr);
}
}
for (label proci=0; proci<nProcs; proci++)
{
Info<< "Reading mesh to add from "
<< databases[proci].caseName()
<< " for time = " << databases[proci].timeName()
<< nl << nl << endl;
}
// See if any points on the mastermesh have become connected
// because of connections through processor meshes.
mergeSharedPoints(mergeDist, masterMesh[0], pointProcAddressing);
// Save some properties on the reconstructed mesh
masterInternalFaces = masterMesh[0].nInternalFaces();
masterOwner = masterMesh[0].faceOwner();
Info<< "\nWriting merged mesh to "
<< runTime.path()/runTime.timeName()
<< nl << endl;
if (!masterMesh[0].write())
{
FatalErrorInFunction
<< "Failed writing polyMesh."
<< exit(FatalError);
}
if (writeCellDist)
{
writeCellDistance
(
runTime,
masterMesh[0],
cellProcAddressing
);
}
}
// Write the addressing
Info<< "Reconstructing the addressing from the processor meshes"
<< " to the newly reconstructed mesh" << nl << endl;
forAll(databases, proci)
{
Info<< "Reading processor " << proci << " mesh from "
<< databases[proci].caseName() << endl;
polyMesh procMesh
(
IOobject
(
regionName,
databases[proci].timeName(),
databases[proci]
)
);
// From processor point to reconstructed mesh point
Info<< "Writing pointProcAddressing to "
<< databases[proci].caseName()
/procMesh.facesInstance()
/polyMesh::meshSubDir
<< endl;
labelIOList
(
IOobject
(
"pointProcAddressing",
procMesh.facesInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false // Do not register
),
pointProcAddressing[proci]
).write();
// From processor face to reconstructed mesh face
Info<< "Writing faceProcAddressing to "
<< databases[proci].caseName()
/procMesh.facesInstance()
/polyMesh::meshSubDir
<< endl;
labelIOList faceProcAddr
(
IOobject
(
"faceProcAddressing",
procMesh.facesInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false // Do not register
),
faceProcAddressing[proci]
);
// Now add turning index to faceProcAddressing.
// See reconstructPar for meaning of turning index.
forAll(faceProcAddr, procFacei)
{
const label masterFacei = faceProcAddr[procFacei];
if
(
!procMesh.isInternalFace(procFacei)
&& masterFacei < masterInternalFaces
)
{
// proc face is now external but used to be internal face.
// Check if we have owner or neighbour.
label procOwn = procMesh.faceOwner()[procFacei];
label masterOwn = masterOwner[masterFacei];
if (cellProcAddressing[proci][procOwn] == masterOwn)
{
// No turning. Offset by 1.
faceProcAddr[procFacei]++;
}
else
{
// Turned face.
faceProcAddr[procFacei] =
-1 - faceProcAddr[procFacei];
}
}
else
{
// No turning. Offset by 1.
faceProcAddr[procFacei]++;
}
}
faceProcAddr.write();
// From processor cell to reconstructed mesh cell
Info<< "Writing cellProcAddressing to "
<< databases[proci].caseName()
/procMesh.facesInstance()
/polyMesh::meshSubDir
<< endl;
labelIOList
(
IOobject
(
"cellProcAddressing",
procMesh.facesInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false // Do not register
),
cellProcAddressing[proci]
).write();
// From processor patch to reconstructed mesh patch
Info<< "Writing boundaryProcAddressing to "
<< databases[proci].caseName()
/procMesh.facesInstance()
/polyMesh::meshSubDir
<< endl;
labelIOList
(
IOobject
(
"boundaryProcAddressing",
procMesh.facesInstance(),
polyMesh::meshSubDir,
procMesh,
IOobject::NO_READ,
IOobject::NO_WRITE,
false // Do not register
),
boundaryProcAddressing[proci]
).write();
Info<< endl;
}
}
Info<< "End.\n" << endl;
return 0;
}
// ************************************************************************* //