Hello Federico, I was having the same problem with binary black holes evolutions with IllinoisGRMHD. In my case, the "exit(1)" statement in harm_primitives_lowlevel.C in the piece of code you copied above silently killed my runs when the Font fix failed, pretty much as in your run. Just commenting it out solved that problem for me, so you can give it a try if you want.
On Thu, Sep 22, 2022, 07:37 Federico Cattorini f.cattorini@campus.unimib.it wrote:
Hello Erik,
Thanks for your reply.
I followed your suggestions and found two strings that may be indicative of what's going on. In the standard outputs 'CCTK_Proc14.out' and 'CCTK_Proc15.out' the last lines read
INFO (IllinoisGRMHD): Font fix failed! INFO (IllinoisGRMHD): i,j,k = 67 63 16, stats.failure_checker = 0 x,y,z = 3.392857e+00 8.892857e+00 2.392857e+00 , index=111739 st_i = -1.002115e+08 2.298583e+08 -1.221746e+08, rhostar = 1.573103e+02, Bi = -1.064528e+03 1.120144e+03 2.972675e+03, gij = 6.643816e+00 5.521615e-01 4.380688e-01 7.244355e+00 -1.685406e-03 6.830374e+00, Psi6 = 1.803534e+01
I assume this means that there are issues in the con2prim of IGM. That INFO is printed by harm_primitives_lowlevel.C:
// Use the new Font fix subroutineint font_fix_applied=0; if(check!=0) { font_fix_applied=1; CCTK_REAL u_xl=1e100, u_yl=1e100, u_zl=1e100; // Set to insanevalues to ensure they are overwritten. if (gamma_equals2==1) { check = font_fix_gamma_equals2(u_xl,u_yl,u_zl,CONSERVS,PRIMS,METRIC_PHYS,METRIC_LAP_PSI4,eos); } else { check = font_fix_general_gamma(u_xl,u_yl,u_zl,CONSERVS,PRIMS,METRIC_PHYS,METRIC_LAP_PSI4,eos); } //Translate to HARM primitive now: prim[UTCON1] = METRIC_PHYS[GUPXX]*u_xl + METRIC_PHYS[GUPXY]*u_yl + METRIC_PHYS[GUPXZ]*u_zl; prim[UTCON2] = METRIC_PHYS[GUPXY]*u_xl + METRIC_PHYS[GUPYY]*u_yl + METRIC_PHYS[GUPYZ]*u_zl; prim[UTCON3] = METRIC_PHYS[GUPXZ]*u_xl + METRIC_PHYS[GUPYZ]*u_yl + METRIC_PHYS[GUPZZ]*u_zl; if (check==1) { CCTK_VInfo(CCTK_THORNSTRING,"Font fix failed!"); CCTK_VInfo(CCTK_THORNSTRING,"i,j,k = %d %d %d, stats.failure_checker = %d x,y,z = %e %e %e , index=%d st_i = %e %e %e, rhostar = %e, Bi = %e %e %e, gij = %e %e %e %e %e %e, Psi6 = %e",i,j,k,stats.failure_checker,X[index],Y[index],Z[index],index,mhd_st_x_orig,mhd_st_y_orig,mhd_st_z_orig,rho_star_orig,PRIMS[BX_CENTER],PRIMS[BY_CENTER],PRIMS[BZ_CENTER],METRIC_PHYS[GXX],METRIC_PHYS[GXY],METRIC_PHYS[GXZ],METRIC_PHYS[GYY],METRIC_PHYS[GYZ],METRIC_PHYS[GZZ],METRIC_LAP_PSI4[PSI6]); exit(1); // Let's exit instead of printing potentially GBs of log files. Uncomment if you really want to deal with a mess. } } stats.failure_checker+=font_fix_applied*10000; stats.font_fixed=font_fix_applied;
Can I do anything that may help pinpoint the cause of this error?
Thanks in advance,
Federico
Il giorno gio 15 set 2022 alle ore 15:30 Erik Schnetter < schnetter@gmail.com> ha scritto:
Federico
Thanks for including the output, that is helpful.
There are parameters "Carpet::verbose" and "Carpet::veryverbose". You can set them to "yes" and recover from a checkpoint. This gives more information about what the code is doing, and thus where it crashes.
The output you attached is only from the first MPI process. Other processes' output might contain a clue. You can add the command line option "-roe" to Cactus when you run the simulation. This will collect output from all processes.
-erik
On Thu, Sep 15, 2022 at 9:20 AM Federico Cattorini f.cattorini@campus.unimib.it wrote:
Hello everyone,
I am experiencing some issue in a number of GRMHD simulations of black
hole binaries employing IllinoisGRMHD.
As an example, I will write about an unequal-mass BHB configuration
(with q = 2) that I'm running.
After approximately ten orbits, the run stops with no error codes or
any other message that could help me identify the issue. The last lines of the standard output are
INFO (IllinoisGRMHD): ***** Iter. # 353949, Lev: 9, Integrating to
time: 3.160260e+03 *****
INFO (IllinoisGRMHD): C2P: Lev: 9 NumPts= 569160 | Fixes: Font= 393 VL=
179 rho*= 2 | Failures: 0 InHoriz= 0 / 0 | Error: 7.124e-02, ErrDenom: 4.838e+13 | 4.51 iters/gridpt
INFO (IllinoisGRMHD): ***** Iter. # 353949, Lev: 9, Integrating to
time: 3.160269e+03 *****
Simfactory Done at date: gio 04 ago 2022 11:43:01 CEST
I tried restarting my simulation from the latest checkpoint, but the
same sudden stop occurred at the same timestep.
At first, I thought about some problem with IGM. The last INFO is
printed by IllinoisGRMHD_driver_evaluate_MHD_rhs.C, so I put some prints in it to identify the spot where the error occurs.
Unfortunately, I drew a blank, since the stop seems to occur just after
the end of IllinoisGRMHD_driver_evaluate_MHD_rhs:
INFO (IllinoisGRMHD): ***** line 52: entering
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
INFO (IllinoisGRMHD): ***** Iter. # 353949, Lev: 10, Integrating to
time: 3.160251e+03 *****
INFO (IllinoisGRMHD): ***** line 100:
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
INFO (IllinoisGRMHD): ***** line 204: just before
reconstruct_set_of_prims_PPM *****
INFO (IllinoisGRMHD): ***** DEBUG END of
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
Simfactory Done at date: gio 04 ago 2022 19:44:55 CEST
I tried to restart the simulation and run it on pure MPI. It ran for a
few more iterations, then stopped as well:
INFO (IllinoisGRMHD): ***** line 52: entering
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
INFO (IllinoisGRMHD): ***** Iter. # 353565, Lev: 10, Integrating to
time: 3.156831e+03 *****
INFO (IllinoisGRMHD): ***** line 100:
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
INFO (IllinoisGRMHD): ***** line 204: just before
reconstruct_set_of_prims_PPM *****
INFO (IllinoisGRMHD): ***** DEBUG END of
IllinoisGRMHD_driver_evaluate_MHD_rhs *****
Simfactory Done at date: ven 05 ago 2022 19:00:13 CEST
The simulation setup is as follows:
Allocated: Nodes: 10 Cores per node: 48 SLURM setting SLURM_NNODES : 10 SLURM_NPROCS : 20 SLURM_NTASKS : 20 SLURM_CPUS_ON_NODE : 48 SLURM_CPUS_PER_TASK : 24 SLURM_TASKS_PER_NODE: 2(x10) Running: MPI processes: 20 OpenMP threads per process: 24 MPI processes per node: 2.0 OpenMP threads per core: 1.0 OpenMP threads per node: 48
while the pure-MPI setup is
Allocated: Nodes: 10 Cores per node: 48 SLURM setting SLURM_NNODES : 10 SLURM_NPROCS : 480 SLURM_NTASKS : 480 SLURM_CPUS_ON_NODE : 48 SLURM_CPUS_PER_TASK : 1 SLURM_TASKS_PER_NODE: 48(x10) Running: MPI processes: 480 OpenMP threads per process: 1 MPI processes per node: 48.0 OpenMP threads per core: 1.0 OpenMP threads per node: 48
I am using The Lorentz version of ET.
I've had this issue for two binary BH simulations, both unequal-mass
with q = 2. My colleague Giacomo Fedrigo experienced the same problem running an equal-mass simulation.
I attach the q = 2 (s_UUmis_Q2) parameter file and the ET config-info
file. Also, I attach the st. error and output of my q = 2 run and of Giacomo's run (b1_UUmis_a12b_pol3_r56_gauss_9). The st. outputs were cut for readability reasons.
Can someone please help me with this?
Thanks in advance,
Federico _______________________________________________ Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
-- Erik Schnetter schnetter@gmail.com http://www.perimeterinstitute.ca/personal/eschnetter/
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users