Fluids

Fluids

Overset mesh adaptation: segmentation error v2022R1/2

    • Axel Bral
      Subscriber

      In my case with overset and the flow-blocking gap model, I want to eliminate the few remaining orphans. Since the orphan cells have a rather high aspect ratio (10), I would like to split the orphans themselves as some of the partitions would then be taken up by the gap definition, while the others might find suitable donors after adaptation. Increasing the gap threshold is not an option, since it then would become excessevely large. Increasing the global cell count to reduce the aspect ratio would lead to a too high cell count. As a consequence, adapting the orphans is the preferred choice. Therefore, I took following steps:

      /solve/cell-registers/add "orphans" type field-value field overset-cell-type option less-than -0.1 q q
      /mesh/adapt/cell-registers/refine "orphans"

      which then lead to a segmentation fault:

      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_host/fluent.22.2.0': double free or corruption (!prev): 0x00000000054bd400 ***

      ==============================================================================

      Node 2: Process 258855: Received signal SIGSEGV.

      ==============================================================================
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x000000000790e380 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x0000000006f1a6f0 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': free(): corrupted unsorted chunks: 0x0000000005c71f20 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': free(): corrupted unsorted chunks: 0x0000000006da2f40 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x0000000007162a70 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': free(): corrupted unsorted chunks: 0x00000000065c0f40 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': free(): corrupted unsorted chunks: 0x00000000076f4f40 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x00000000078461d0 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': free(): corrupted unsorted chunks: 0x00000000065bbf20 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x0000000007f01e80 ***
      *** Error in `/apps/CO7/ANSYS/2022R2/ansys_inc/v222/fluent/fluent22.2.0/lnamd64/3ddp_node/fluent_mpi.22.2.0': double free or corruption (!prev): 0x0000000006b8d430 ***

      I also noticed that any type of mesh refinement using the /mesh/refine tab leads to segmentation faults as soon as overset meshes are involved. Therefore, I also tried the overset mesh adaptation, where the cells of other zones overlapping with the orphans are marked (orphan adaptation), though this is not the preferred behavior for my case. The steps are listed here below:

      /file/read-case-data "test_03-04-14-15.cas.h5"
      /define/overset-interfaces/adapt/set/adaption-sweeps 1
      /define/overset-interfaces/adapt/set/automatic no
      /define/overset-interfaces/adapt/set/buffer 0
      /define/overset-interfaces/adapt/set/mark-coarsening? no
      /define/overset-interfaces/adapt/set/mark-fixed-orphans? no
      /define/overset-interfaces/adapt/set/mark-orphans? yes
      /define/overset-interfaces/adapt/set/mark-size? no
      /define/overset-interfaces/adapt/set/maximum-refinement-level 1
      /define/overset-interfaces/adapt/mark-adaption
      /define/overset-interfaces/adapt/adapt-mesh

      Consequently, I once more receive a segmentation fault:

      > /define/overset-interfaces/adapt/adapt-mesh

      ===============Message from the Cortex Process================================

      Fatal error in one of the compute processes.

      ==============================================================================

      Output of the problematic compute node:

       myid (33): Fatal signal raised sig = SIGSEGV
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0() [0x2a69e8c]
       /lib64/libpthread.so.0(+0x12ce0) [0x146e9c543ce0]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(_ZN5adapt6fluent4Mesh7addCellINS_15PairListAdaptorINS_9SmallListINS_10StaticListINS0_10FluentFaceELm2EEELm10EEEEEEENS0_10FluentCellERKSA_RKT_iilb+0x254) [0x19afe74]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(_ZN5adapt14AdaptionEngineINS_6fluent4MeshEE9refineHexERKNS1_10FluentCellE+0x148f) [0x19ae0df]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(_ZN5adapt14AdaptionEngineINS_6fluent4MeshEE24performRefinementAtLevelEi+0x859) [0x1990259]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(AdaptFluentMesh+0x18bf) [0x1986ecf]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(DynamicAdaptPolyhedralMesh+0x84) [0x15ed9f4]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(Overset_Adapt_Mesh+0x601) [0x1640fd1]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0() [0x65f7c5]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(eval+0x497) [0x2a84317]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(PRF_Command_Start+0x133c) [0x6db6cc]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(PRF_Node_repl+0x2e) [0x6dc98e]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(fluent_main+0x3f9) [0x6f4279]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0(main+0x59) [0x70a869]
       /lib64/libc.so.6(__libc_start_main+0xf3) [0x146e8ba25cf3]
       /apps/gent/RHEL8/cascadelake-ib/software/FLUENT/2022R1/v221/fluent/fluent22.1.0/lnamd64/3ddp_node/fluent_mpi.22.1.0() [0x5d74a1]

      Even though the case is rather large, a lack of memory is certainly not the cause since only 60 GB of the available 738 GB of the cluster was used during this job. 

       

      Is this a known bug? Any help is welcome.

    • C N
      Ansys Employee

      Hello,

      I recommend instead of UDF for overset meshes, it is recommended that the gaps between walls are resolved by a minimum of 4 cells, in order to create a valid overset interface and prevent the creation of orphan cells. You can specify that the cells in gaps with a resolution that is less than this recommendation are marked, so that any subsequent overset adaption (manual or automatic) increases the resolution to 4 or more cells.For further understanding kindly refer this link 

      3.4. Overset Gap Adaption (ansys.com)

      If your not able to access this link kindly refer to this link Using Help with links (ansys.com)

      Thanks

Viewing 1 reply thread
  • You must be logged in to reply to this topic.