View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0004118 | OpenFOAM | Bug | public | 2024-07-27 13:45 | 2024-07-27 14:25 |
| Reporter | cgoessni | Assigned To | henry | ||
| Priority | normal | Severity | minor | Reproducibility | sometimes |
| Status | closed | Resolution | no change required | ||
| Product Version | 12 | ||||
| Fixed in Version | dev | ||||
| Summary | 0004118: XiFluid ignition causes memory corruption with meshToMesh under certain circumstances | ||||
| Description | When running ignition in XiFluid v12, and the ignition extends through a meshToMesh mapping, the ignition can cause memory corruption when the ignition cells switch from one processor to another. This is due to not setting the cells_ and volumes_ arrays to zero before the return: // Bit tricky: generate C and V before shortcutting if cannot find // cell locally. mesh.C generation uses parallel communication. const volVectorField& centres = mesh.C(); const scalarField& vols = mesh.V(); label ignCell = mesh.findCell(location_); if (ignCell == -1) * { * return; } We had an engine case that sometimes crashed with certain model combinations, and the crashes were always random (free(): invalid next size, double-linked list corruption, segfault, ...). If we set the size of cells_ and volumes_ to zero before return, all crashes go away: diff --git a/src/thermophysicalModels/ignition/ignitionSite.C b/src/thermophysicalModels/ignition/ignitionSite.C index 320467f..7401146 100644 --- a/src/thermophysicalModels/ignition/ignitionSite.C +++ b/src/thermophysicalModels/ignition/ignitionSite.C @@ -39,6 +39,9 @@ void Foam::ignitionSite::findIgnitionCells(const fvMesh& mesh) label ignCell = mesh.findCell(location_); if (ignCell == -1) { + cells_.setSize(0); + cellVolumes_.setSize(0); + return; } | ||||
| Steps To Reproduce | Very difficult to reproduce: the ignition cells have to switch from procA to procB (as with our engine case), making cells_ and volumes_ invalid but still with size > 0 on procA. But even under these conditions, some model combinations worked sometimes, sometimes we saw deadlocks and for certain model combinations, we saw random (i.e., always different) crashes. | ||||
| Additional Information | We know that in -dev, this is fixed with the fundamentally different approach. It would still be nice to have this small bugfix in v12, if possible? | ||||
| Tags | No tags attached. | ||||
|
|
I have made lots of improvements to XiFluid in OpenFOAM-dev, in particular with respect to ignition which had many issues. I recommend that you use the latest XiFluid solver and if you are not able to run OpenFOAM-dev but can compile changes to OpenFOAM-12 you could replace the XiFluid in OpenFOAM-12 with the current version from OpenFOAM-dev. |
|
|
Already resolved in OpenFOAM-dev |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2024-07-27 13:45 | cgoessni | New Issue | |
| 2024-07-27 14:24 | henry | Note Added: 0013324 | |
| 2024-07-27 14:25 | henry | Assigned To | => henry |
| 2024-07-27 14:25 | henry | Status | new => closed |
| 2024-07-27 14:25 | henry | Resolution | open => no change required |
| 2024-07-27 14:25 | henry | Fixed in Version | => dev |
| 2024-07-27 14:25 | henry | Note Added: 0013325 |
