Open nicolasaunai opened 3 months ago
changing the script to 32/12 cells, and running with 20 cores...
the other ranks are all this, in some form, some cut off for unknown reasons
same script as initially ran but on 80 cores still failed with similar errors. Re-running on 40 cores.
The issue is not the number of cores (which was to high anyway), but probably that this setup leads to lots of magnetic fluctuations in the low density region, thereby ending up with the whole domain refined. Why would that lead to a segmentation fault when rebalancing L0 I don't know though....
Higher threshold led to a more local refinement and no more error. See #879 to play with thresholds
Ran with 1eff256247
there's no 27.out, no 23.out, ranks we find in the like of those which segfault below. There's a 26,25,28, however...
run065a.txt
From above output file:
The fact that it happens exactly after t=1 and I have regriding L0 setup with 1000dt cadence let me thinks something is wrong there.