Destroy object the PETSc garbage collector would have issue with#5235
Destroy object the PETSc garbage collector would have issue with#5235UZerbinati wants to merge 1 commit into
Conversation
|
Please add parallel tests |
|
This shouldn't be hanging... the garbage collector should never do this |
What do you mean ? ngsPETSc and netgen interface is already tested in parallel, no ? |
This does not seem to be the case for the adaptive refinement code in Netgen. However we're going to replace this ad-hoc renumbering hell with DMAdaptLabel in #5215 |
|
Was this happening reliably or stochastically? Others have also random observed garbage collector hangs and I would like to resolve it. |
How old are those reports? I fixed a very common hanging issue that was triggered by repetedly calling multigrid not too long ago https://gitlab.com/petsc/petsc/-/merge_requests/8694 |
Looking forward to that :) |
Happened consistently on my machine and on @ThighamOxford machine consistently. Everything was up to date current main. |
That's awesome. If I find the time I really need to dig into what's causing this. |
|
I still believe is not a bad idea to destroy stuff we will not use anyway, no ? Or is this not standard practice in Firedrake ? |
In theory it's a completely pointless thing to do because Python will destroy things for us. We try to avoid C-style coding in favour of being Pythonic. This should be added to our style guide. However, it doesn't do any harm and is clearly exposing an issue with the garbage collector. I am happy to merge this if @pbrubeck is. |
I think we should give it another go with Claude with this in mind, it could be very helpful to get to the bottom of this. |
| section0, mvec0 = self.topology_dm.distributeField(sfBCInv, | ||
| self._cell_numbering, | ||
| mvec) | ||
| mark_np = mvec0.getArray().copy() |
There was a problem hiding this comment.
Maybe this copy is all we needed?
3 participants
PETSc Garbage collector was hanging on my machine when running test locally :)